Care loopd - #4
Open
Jacobjeevan wants to merge 27 commits into
Open
Conversation
Jacobjeevan
commented
Jul 13, 2026
Jacobjeevan
commented
Contributor
- Added care loop and loop doctor skills
Optional pre-Step-1 enrichment (TicketFetcher port) that folds a Jira ticket's description — ADF flattened, with inline [image: NAME] markers keeping each image welded to the paragraph it illustrates — plus its image attachments into the planner input. Attachments ride both planner phases as opencode file parts (the image->model transport probed on Copilot). Cached under runDir + resume-safe; degrades to the raw kickoff task when unconfigured or on fetch failure, so the default deployment (no JIRA_* env) is unchanged. Live-verified end to end against CARE-298: classic Basic auth, issue fetch, ADF flatten with document-ordered attachments, authenticated image download, and the image reaching the model. smoke:jira / probe:image cover the live paths. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ndness, specs-owed) Fixes proposed by the end-of-run doctor across ~6 runs, collated in care-loop-doctor/proposals/000-COLLATION-2026-07-28.md, plus the doctor --report mode that produced them: - [§A] triager: skip already-resolved threads entirely — emit no verdict at all (not even decline), so a round stops burning spawns re-litigating threads closed rounds ago. Only genuinely-new content on a re-opened thread earns a fresh verdict. - [§E.1] defaultDiffOf: include untracked new files (synthesized add-diff), since `git add -A` only runs at step 5 — a brand-new *.spec.ts was invisible to the reviewer/test-grader meant to grade it. - [§E.2] test-grader: raise a `specs_owed` advisory (listing the unasserted criteria) when the plan declared a Test-surface contract but zero specs were delivered, instead of a silent no_specs pass. Advisory only — never hard-gates. - doctor: `--report` mode writes one proposal doc per run instead of opening a PR.
care-diff-review Step 2 is now the `care-intent` skill, so it can run as a standalone maker-tier role (the salvage reconstruction spawn) and be graded by care-evals. care-diff-review keeps the same prompt: its `default` methodology region splits into `agreement` + `findings`, and `reviewerMethodology` composes agreement → care-intent → findings in file order. Byte-identity of the 4a reviewer prompt is pinned by two golden fixtures (tsx / no-tsx), plus guards that the answer-key rule stays out and the Reference block stays in. care-evals: `care-intent` grades on the care-review path (recall/FP), with its own prompt builder and a first task (it-01-age-short-misread). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
… open PR Salvaging an old PR is not a new flow, it is the existing loop entered late: Steps 1-5 already happened and their outputs sit on the PR, so what remains is 6a/6b/7, which runCiRounds already drives. - adopt.ts: reconstruct the PR's intent BLIND (the description is never fed to the model), confirm at one human gate with a description-vs-diff divergence check, bootstrap the run dir + a synthetic journal that projects to mode "ci". Adopt backdates a synthetic push so round 1's poll baseline counts the PR's existing reviews as arrived, instead of waiting forever on an unchanged head. - salvage-gate-terminal.ts: the single human gate for the above. - reviewers.ts + feedback.ts: opt-in digest extensions — review summary bodies, unanchored path:line findings, reviewer attribution via pull_request_review_id, per-source trim budgets so trusted reviewers' fix suggestions survive trimming. Every option defaults off, so the loop's digest stays byte-identical (pinned by a snapshot test) and makes no extra API call. - github.ts: carry PR body/baseRef and review id/body. - cli.ts: `--pr <n>` wiring, worktree checkout, then hand off to cmdResume. - skills-opencode.ts: opencodeIntentReconstructor (care-intent, maker tier). 266 tests green; tsc clean. Live dogfood on a real PR still pending. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
…ion settings Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.